AI and the World's Hard Problems · Physics · Nature Machine Intelligence · 2026-10-20
Lagrangian–Eulerian learning of flow field and trajectories with TrajectoryFlowNet
Predicting particle transport in complex flows usually requires deep physical insight and high computational cost. TrajectoryFlowNet, a Lagrangian–Eulerian physics-informed neural network, predicts flow fields and long-range particle trajectories from scarce trajectory measurements. It was validated on numerical cases and on aortic and ventricle blood flows, showing it can track particles over long ranges and build accurate flow fields.